Tag-Align: Do you actually need the sequence part

[dagarfield]

Hi all,

We're gearing up to try your pipeline on some in house generated scATAC-seq data (currently stored as an ArchR project). From what I understand, we need to translate our data (which is now essentially in fragment files) to Tag-Align format. This is straight forward except that we haven't recorded the sequence.

So -- in your pipeline, is the sequence field of the TagAlign necessary?

We may have other questions as we go.....

[mayasheth]

Hi, thanks for the question! You are correct that to use the ENCODE-rE2G model, you would need to convert your fragment files to tagAlign files. When you refer to the "sequence," do you mean the fourth column in the fragment file? If so this is actually the cell barcode, and is not necessary for the tagAlign conversion.

We actually have a model adapted from this one called scE2G that is built specifically for single-cell data (both multiome and ATAC only), available here: https://github.com/EngreitzLab/scE2G

This pipeline takes the fragment files directly as input and might be a better choice for you. Happy to advise further!

[dagarfield]

Oh, I'll take a look!

[dagarfield]

In a nutshell, are there fundamental changes in the pipeline? Or more some wrappers to generate pseudo-bulks per cell type?

[mayasheth]

The models have been fine-tuned for scATAC activity measurements and we apply additional normalization to account for variation in sequencing depth, etc. Given high quality/depth data, we see similar performance for ENCODE-rE2G and scE2G. If you are working with small cell clusters (less than 5M fragments or so), I would recommend scE2G. If you are aiming to integrate other bulk datasets (H3K27ac, Hi-C), I would stick with ENCODE-rE2G.